Earnings Precision and the Relations between Earnings and Returns

Abstract

Formal and heuristic valuation models suggest that changes in firm value associated with a revision in expected earnings should be a multiple on the order of typical earnings capitalization factors between 10 and 30. However, empirical estimates of the earnings response coefficient (ERC) have usually been in a range an order of magnitude lower, between 1 and 3 (Kothari 2001). This paper uses a simple Bayesian model to integrate previous theoretical and empirical results and bridge this large gap. In the Bayesian model, low precision implies a low weight on new earnings information. As a result, the observable earnings surprise is much larger than the unobservable earnings revision, which in turn implies a smaller coefficient on earnings surprise than on the revision in expected earnings. Although the model uses statistical measures, the Bayesian concept of precision closely parallels the accounting concept of earnings quality, i.e., the extent to which current earnings is informative about expected future earnings. Precision summarizes the effects of multiple determinants of earnings quality.Consistent with the model, two proxies for precision, forecast dispersion and absolute magnitude of earnings surprise, explain a broad empirical range of coefficients on earnings surprise ranging from near 0 up to 30. We reconcile the surprisingly large proportion of observations falling in the upper end of the range with the surprisingly low estimated ERCs typically reported in previous research. Finally, we demonstrate in the appendix how the precision proxies can be used to design tests that focus on the majority of observations with larger surprise coefficients and thereby allow researchers to detect the effect of determinants of surprise coefficients that would otherwise be empirically undetectable.